1 Introduction
1.1 Results
2 First- and second-order difference-constrained vectors
2.1 Overview of the algorithm
2.2 Transformation from state Pi-1 to Pi
2.3 Algorithm
3 Conclusion
A Simple greedy algorithm for convex regression
B Proof of Lemma 2.8
C Complete algorithm
C.1 Implicitly computing the Pi
C.2 Backtracing
C.3 Analysis
D Generalization to DAGs is hard
